Resource scoring model training method and device, electronic equipment and storage medium

Abstract

The invention provides a resource scoring model training method and device, electronic equipment and a storage medium, relates to the technical field of artificial intelligence, in particular to the technical fields of deep learning, large models and the like, and can be used for application scenes such as recommendation systems, content screening and sorting and the like. According to the specific implementation scheme, the method comprises the steps of obtaining a first resource feature of a pushed resource in a first time period and a second resource feature of a to-be-pushed resource in a second time period; acquiring behavior characteristics of the target object to the pushed resources and basic characteristics of the target object in the first time period; and inputting the first resource feature, the second resource feature, the behavior feature and the basic feature into a resource scoring model, so that the resource scoring model at least outputs a first class score of the to-be-pushed resource. According to the scheme, by introducing the resource scoring model and comprehensively considering various feature information, accurate scoring and recommendation of the to-be-pushed resources can be realized, so that continuous consumption of the resources by the user can be effectively driven.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, particularly to technologies such as deep learning and large models, and can be used in application scenarios such as recommendation systems, content screening and sorting. Specifically, it relates to a method, device, electronic device, and storage medium for training a resource scoring model. Background Art

[0002] With the continuous consumption of users in the new fusion scenario, their satisfaction may gradually decline, and their interest in the recommended content will also gradually weaken. Therefore, the recommendation effect of the recommendation system needs to be strengthened. Summary of the Invention

[0003] The present disclosure provides a method, device, electronic device, and storage medium for training a resource scoring model.

[0004] According to a first aspect of the present disclosure, there is provided a method for training a resource scoring model, including: obtaining first resource features of the resources that have been pushed within a first time period and second resource features of the resources to be pushed within a second time period, where the second time period is the next time period of the first time period; obtaining the behavioral features and basic features of the target object for the resources that have been pushed within the first time period; inputting the first resource features, second resource features, behavioral features, and basic features into the resource scoring model, so that the resource scoring model outputs at least a first type of score for the resources to be pushed; where the first type of score is used to predict the relative change in the behavioral performance of the target object for the resources to be pushed within the second time period compared to the behavioral performance for the resources that have been pushed within the first time period.

[0005] According to a second aspect of the present disclosure, there is provided a resource scoring method, including: obtaining the behavioral features, basic features, first resource features of the resources that have been pushed, and second resource features of the resources to be pushed within the next time period of the target object within the current time period; inputting the behavioral features, basic features, first resource features, and second resource features into the resource scoring model to obtain the first type of score output by the resource scoring model; where the first type of score is used to predict the relative change in the behavioral performance of the target object for the resources to be pushed within the next time period compared to the behavioral performance of the target object for the resources that have been pushed within the current time period; where the resource scoring model is trained by the method of the first aspect or any optional implementation manner of the first aspect.

[0006] According to a third aspect of the present disclosure, there is provided a training device for a resource scoring model, including: a resource feature acquisition module, configured to acquire first resource features of the resources that have been pushed within a first time period and second resource features of the resources to be pushed within a second time period, where the second time period is the next time period of the first time period; a user feature acquisition module, configured to acquire behavioral features of the target object for the resources that have been pushed and basic features of the target object within the first time period; a scoring model training module, configured to input the first resource features, the second resource features, the behavioral features, and the basic features into the resource scoring model, so that the resource scoring model outputs at least a first type of score for the resources to be pushed; wherein the first type of score is used to predict the relative change in the behavioral performance of the target object for the resources to be pushed within the second time period compared to the behavioral performance for the resources that have been pushed within the first time period.

[0007] According to a fourth aspect of the present disclosure, there is provided a resource scoring device, including: a feature acquisition module, configured to acquire behavioral features of the target object, basic features, first resource features of the resources that have been pushed, and second resource features of the resources to be pushed within the next time period within the current time period; a resource scoring module, configured to input the behavioral features, the basic features, the first resource features, and the second resource features into the resource scoring model to obtain a first type of score output by the resource scoring model; wherein the first type of score is used to predict the relative change in the behavioral performance of the target object for the resources to be pushed within the next time period compared to the behavioral performance of the target object for the resources that have been pushed within the current time period; wherein the resource scoring model is trained by the method of the first aspect or any optional implementation manner of the first aspect.

[0008] According to a fifth aspect of the present disclosure, there is provided an electronic device, including:

[0009] at least one processor; and

[0010] a memory communicatively connected to the at least one processor; wherein,

[0011] the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute any method in the embodiments of the present disclosure.

[0012] According to a sixth aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute any method in the embodiments of the present disclosure.

[0013] According to a seventh aspect of the present disclosure, there is provided a computer program product, including a computer program, which when executed by a processor, implements any method in the embodiments of the present disclosure.

[0014] By adopting the solution of the present disclosure, through introducing a resource scoring model and comprehensively considering various feature information, accurate scoring and recommendation of resources to be pushed can be achieved, thereby effectively driving users' continuous consumption of resources.

[0015] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become easily understandable through the following description. Brief Description of the Drawings

[0016] The drawings are used to better understand the solution and do not constitute a limitation to the present disclosure. Among them:

[0017] Figure 1 is a schematic flowchart of a method for training a resource scoring model according to an embodiment of the present disclosure;

[0018] Figure 2 is another schematic flowchart of a method for training a resource scoring model according to an embodiment of the present disclosure;

[0019] Figure 3 is a schematic flowchart of a resource scoring method according to an embodiment of the present disclosure;

[0020] Figure 4 is a schematic structural diagram of a device for training a resource scoring model according to an embodiment of the present disclosure;

[0021] Figure 5 is a schematic structural diagram of a resource scoring device according to an embodiment of the present disclosure;

[0022] Figure 6 is a schematic scenario diagram of a method for training a resource scoring model according to an embodiment of the present disclosure;

[0023] Figure 7 is a schematic scenario diagram of a resource scoring method according to an embodiment of the present disclosure;

[0024] Figure 8 is a schematic structural diagram of an electronic device for implementing the method for training a resource scoring model and / or the resource scoring method according to an embodiment of the present disclosure. Detailed Embodiments

[0025] The following makes an explanation of exemplary embodiments of the present disclosure with reference to the drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of the present disclosure. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted below.

[0026] As used herein, the term "and / or" merely describes an associative relationship between associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist simultaneously, and B exists alone. As used herein, the term "at least one" means any one of a plurality or any combination of at least two of a plurality. For example, including at least one of A, B, and C may represent any one or more elements selected from the set composed of A, B, and C. As used herein, the terms "first" and "second" refer to and distinguish multiple similar technical terms, and do not mean to limit the order or limit to only two. For example, the first feature and the second feature refer to two categories / two features. The first feature may be one or more, and the second feature may also be one or more.

[0027] In addition, for a better illustration of the present disclosure, numerous specific details are given in the following specific embodiments. Those skilled in the art should understand that the present disclosure can also be implemented without certain specific details. In some instances, methods, means, elements, and circuits well-known to those skilled in the art are not described in detail to highlight the gist of the present disclosure.

[0028] Before introducing the technical solutions of the embodiments of the present disclosure, further explanations will be made on the technical terms that may be used in the present disclosure:

[0029] New fusion scenario: A scenario of deep combination in multiple fields and dimensions. In this scenario, different types of resources, services, systems, or experiences are integrated to meet the more personalized and complex needs of users. Compared with traditional scenarios, the new fusion scenario pays more attention to cross-border integration and dynamic optimization, and often includes the comprehensive application of multiple business models, technical means, or content forms. Exemplarily, the new fusion scenario may refer to a scenario that can distribute resources such as short videos, small videos, and graphic dynamics, and form an overall resource to build an immersive scenario for users.

[0030] In related technologies, as users continuously consume in the new fusion scenario, the degree of user dissatisfaction gradually increases, and users will gradually become less and less interested. In the current recommendation system, the main means to solve the user satisfaction problem is to establish a model of whether the user is satisfied with the current candidate resource, input user historical sequence features, user model features, candidate resource features, etc., and use whether the user is satisfied with the current candidate resource as a label to train a satisfaction model for online satisfaction prediction of candidate resources. However, related technologies only predict whether the user is satisfied with the current resource, do not consider the connection between the resource and the user state, cannot identify which resources can improve the user state, and cannot effectively drive the continuous consumption of resources by users.

[0031] To at least partially address one or more of the above problems and other potential problems, the present disclosure proposes a method for training a resource scoring model. By introducing a resource scoring model and comprehensively considering various feature information, it can achieve accurate scoring and recommendation of resources to be pushed, which helps to recommend resources that better meet the interests and needs of users in the next time period, improve the satisfaction and conversion rate of recommendations, and thus effectively drive users' continuous consumption of resources.

[0032] It should be specifically noted that the collection and use of all features of the present disclosure are based on the explicit consent of users and comply with relevant laws and regulations.

[0033] Embodiments of the present disclosure provide a method for training a resource scoring model. Figure 1 It is a schematic flowchart of the method for training a resource scoring model according to an embodiment of the present disclosure. This method for training a resource scoring model can be applied to a training device for a resource scoring model. The training device for the resource scoring model is located in an electronic device. The electronic device includes but is not limited to a fixed device and / or a mobile device. For example, the fixed device includes but is not limited to a server, and the server can be a cloud server or a general server. For example, the mobile device includes but is not limited to: mobile phones, tablet computers, personal computers, etc. In some possible implementation manners, the method for training a resource scoring model can also be implemented by a processor calling computer-readable instructions stored in a memory. As Figure 1 shown, the method for training a resource scoring model includes:

[0034] S101. Obtain the first resource features of the resources already pushed within the first time period and the second resource features of the resources to be pushed within the second time period, where the second time period is the next time period after the first time period.

[0035] S102. Obtain the behavioral features of the target object for the resources already pushed within the first time period and the basic features of the target object.

[0036] S103. Input the first resource features, the second resource features, the behavioral features, and the basic features into the resource scoring model so that the resource scoring model outputs at least the first type of score for the resources to be pushed; where the first type of score is used to predict the relative change in the behavioral performance of the target object for the resources to be pushed within the second time period compared to the behavioral performance for the resources already pushed within the first time period.

[0037] Here, the target object refers to the user served by the recommendation system, that is, the individual whose behavior the system hopes to predict and for whom resources are recommended.

[0038] Here, the first time period refers to the time period during which resources have been pushed to the target object.

[0039] Here, the first resource feature refers to the attribute features of the resources that have been pushed within the first time period. These features can include the category, keywords, themes, recommended positions, display forms, popularity, release time, etc. of the resources, and are used to describe the specific information of the pushed resources.

[0040] Here, the second time period refers to the time range immediately following the first time period, that is, the time period when the system plans to recommend resources in the future. In particular, the time span of the second time period can be the same as that of the first time period or different from that of the first time period.

[0041] Here, the second resource feature refers to the attribute features of the resources to be pushed within the second time period. These features are similar to the first resource features, but they correspond to the resources that have not been pushed and are planned to be recommended in the future time period.

[0042] In the embodiments of the present disclosure, the first resource features of the resources pushed within the first time period can be obtained through historical records and the metadata of the resources. Exemplarily, historical log records of the pushed resources are usually saved in the system. By querying the logs, the first resource list pushed to the target object within the first time period can be found. Then, according to the resource numbers or identifiers recorded in the first resource list, the corresponding first resource features can be extracted from the resource database or metadata table. The above is only an exemplary illustration and does not limit all possible situations for obtaining the first resource features, but only not exhaustively list them here.

[0043] Similarly, the second resource features of the resources to be pushed within the second time period can be obtained through the candidate resource set and the metadata of the resources. Exemplarily, the system usually generates a candidate set of resources to be pushed for the target user before the second time period. By traversing the candidate set of resources to be pushed, the second resource list to be pushed can be found. Then, according to the resource numbers or identifiers recorded in the second resource list, the corresponding second resource features can be extracted from the resource database or metadata table. The above is only an exemplary illustration and does not limit all possible situations for obtaining the second resource features, but only not exhaustively list them here.

[0044] Here, the behavior feature refers to the interaction behavior of the target object with the pushed resources within the first time period. Exemplarily, whether the user clicks on the recommended content, whether the user views or reads the content, whether the user likes, comments, collects, purchases, forwards, etc. The behavior feature reflects the actual reaction of the user to the pushed resources, also reflects the consumption status of the user for the pushed resources within the first time period, and indirectly reflects the interest and satisfaction of the user in the pushed resources within the first time period.

[0045] Here, the basic features refer to the static or relatively stable feature information of the target object, which is used to describe the user profile attributes. The basic features may include features such as geographical location, interest preferences, device type, etc.

[0046] In the embodiments of the present disclosure, the behavioral features of the target object for the pushed resources in the first time period can be obtained through the user behavior log. Exemplarily, the recommendation system usually records the interaction data of the target object for the pushed resources to form a user behavior log, which includes information such as user account, resource number, interaction type, timestamp, etc. By traversing the user behavior log, all interaction records in the first time period can be filtered out, and then the specific interaction data of the target object for each pushed resource can be extracted. Then, the specific interaction data can be statistically aggregated to extract the behavioral features of the target object. The above is only an exemplary illustration and does not limit all possible situations for obtaining behavioral features, but no exhaustive listing is made here.

[0047] In the embodiments of the present disclosure, the basic features of the target object can be obtained through the user information database. Exemplarily, the recommendation system usually records the basic information of the target object, which may include data such as region, interest tags, content categories of concern, frequently visited pages, etc. The recommendation system maintains a user profile for each target object based on the basic information. By querying the user profile, the basic features of the target object can be extracted. In particular, the basic features of the target object can also be extracted and generated from the historical data of the target object. The above is only an exemplary illustration and does not limit all possible situations for obtaining basic features, but no exhaustive listing is made here.

[0048] Here, the resource scoring model can be selected according to the actual application scenario. Exemplarily, a sequence modeling method such as a Recurrent Neural Network (RNN) can be selected, or other models such as logistic regression and gradient boosting decision tree can be used to implement. The above is only an exemplary illustration and does not limit all possible situations of the resource scoring model, but no exhaustive listing is made here.

[0049] Here, the first type of score is a core result output by the model, which is used to predict the behavioral performance of the target object for the to-be-pushed resources in the second time period. In particular, this score does not simply predict the behavioral performance of the target object, but rather the change relative to the behavioral performance of the target object for the pushed resources in the first time period. Exemplarily, if the score is higher, it means that the target object may be more interested than in the first time period; if the score is lower, it means that the target object's interest may have decreased.

[0050] In the embodiments of the present disclosure, the first resource feature, the second resource feature, the behavioral feature of the target object, and the basic feature can be first integrated into an input vector of the model. At the same time, according to the behavioral performance of the target object in the first time period and the target behavior of the resource to be pushed in the second time period, a target label of the model is constructed, and then training data is constructed. Next, a suitable loss function is selected to optimize the model, and a suitable optimization algorithm is used to adjust the parameters. Finally, the model is trained using the training data. When the training is completed, the first type of score can be output.

[0051] The technical solution of the embodiments of the present disclosure can model the behavioral change relationship between the first time period and the second time period, and can more accurately predict the change trend of the user's consumption behavior in the future time period. Through the fusion of multi-dimensional features, the model can comprehensively consider the complex relationships among the user's historical behavior, user attributes, and resource attributes, and thus can more accurately evaluate which resources are most suitable for the current user state and make a more reliable prediction of the future user behavior. By using the resource scoring model to predict the change of the target object's behavioral performance in the future time period, the recommended resources are not only highly relevant to the user's current interests, but also help to promote the user's more active behavioral performance, which can significantly improve the matching degree between the recommended resources and the user's preferences, thereby enhancing the overall effect of the recommendation system and user satisfaction.

[0052] In some embodiments, the training method of the resource scoring model further includes: expanding the output ability of the resource scoring model so that the resource scoring model additionally outputs a second type of score of the resource to be pushed based on the first resource feature, the second resource feature, the behavioral feature, and the basic feature; wherein the second type of score is used to predict the satisfaction degree of the target object with respect to the resource to be pushed in the second time period.

[0053] Here, the satisfaction degree of the target object with respect to the resource to be pushed in the second time period refers to the subjective preference and experience degree of the target user for the recommended resource in the second time period.

[0054] In the embodiments of the present disclosure, the satisfaction degree can reflect whether the user likes, approves, or feels that the resource is valuable, and is an important indicator for measuring the effect of the recommendation system, and can be evaluated through directions such as user behavior performance, explicit feedback, or system inference.

[0055] In the embodiments of the present disclosure, the resource scoring model can be designed as a multi-task learning model, so that the model can share input features and some intermediate representations, and optimize different tasks in the high-level branches respectively to achieve the prediction tasks of learning behavioral changes and satisfaction degrees respectively.

[0056] Furthermore, the resource scoring model may include: a bottom-layer feature extraction module, a first type of scoring output branch, and a second type of scoring output branch. Among them, the bottom-layer feature extraction module can extract a general representation from the input first resource feature, second resource feature, behavior feature, and basic feature. Exemplarily, the bottom-layer feature extraction module can be a multi-layer neural network or a feature cross module. Among them, the first type of scoring output branch is used to predict the change in the behavior performance of the target object and output the first type of score. Similarly, the second type of scoring output branch is used to predict the satisfaction of the target object and output the second type of score. Exemplarily, the structure of each branch module can be designed according to the task requirements and may include additional fully connected layers, activation functions, etc. The above is only an exemplary illustration and does not limit all possible situations of the resource scoring model, and only exhaustive listing is not done here.

[0057] In the embodiments of the present disclosure, in the process of expanding the output ability of the resource scoring model so that the resource scoring model additionally outputs the second type of score of the resource to be pushed based on the first resource feature, second resource feature, behavior feature, and basic feature, the first resource feature, second resource feature, behavior feature of the target object, and basic feature can be first integrated into the input vector of the model, and at the same time, according to the satisfaction of the target object with the resource to be recommended in the second time period, the target label of the model is constructed, and then the training data is constructed.

[0058] Furthermore, a first loss function is selected according to the task nature of the first type of score, and a second loss function is selected according to the task nature of the satisfaction prediction. Then, the total loss function of the resource scoring model is determined according to the first loss function and the second loss function. Exemplarily, the weighted sum of the first loss function and the second loss function can be used as the total loss function. The above is only an exemplary illustration and does not limit all possible situations of expanding the output ability of the resource scoring model, and only exhaustive listing is not done here.

[0059] Furthermore, the model is trained using the training data and the total loss function. After the training is completed, the resource scoring model can output the first type of score and the second type of score at the same time.

[0060] Thus, by simultaneously outputting the second type of score, in addition to behavioral data, it is possible to further evaluate whether the resources truly match the user's interests and needs during the current time period, forming a comprehensive measure of the matching degree between the target object and the resources. By combining the first type of score and the second type of score output by the resource scoring model, the recommendation system can more accurately screen and rank the resources to be pushed, and give priority to pushing resources that are likely to attract user behavior and improve user satisfaction. By expanding the output ability, the model can utilize the multi-task learning framework to simultaneously learn the correlation between user behavior and satisfaction, thereby enhancing the understanding ability of user and resource characteristics. This learning method of sharing features can help the model more comprehensively capture the complex relationship between users and resources and improve the overall prediction ability.

[0061] In some embodiments, the training method of the resource scoring model further includes: collecting the behavioral feature data of the target object for the pushed resources within multiple time periods; determining the resource performance scores for each brush in the corresponding time period according to the behavioral feature data within each time period; wherein each time period includes at least one brush; dividing the brushes involved in multiple time periods into positive brushes and negative brushes according to the resource performance scores of multiple brushes within multiple time periods; and generating samples for training the resource scoring model according to the positive brushes and negative brushes.

[0062] Here, a brush refers to the operation of a user refreshing the content in the recommendation system each time. Usually, a set of recommended resources will be presented in one brush. In the recommendation system, a brush is an important analysis unit for analyzing user interaction behavior.

[0063] Here, the resource performance score is an index score that can be used to quantify the performance of all resources in a certain brush, reflecting the user's interest or satisfaction with the brush.

[0064] In the embodiments of the present disclosure, in the process of collecting the behavioral feature data of the target object for the pushed resources within multiple time periods, the interaction data between the user and the resources can be collected in real time through the recommendation system behavior log and the buried point technology in advance, and the behavioral data can be organized and stored by time period and brush to ensure that the data is easy to access and analyze. Moreover, feature extraction can be performed on the behavioral data, and features such as clicks, dwell time, and skip rate can be extracted from the behavioral data to provide a basis for subsequent model training. Exemplarily, for a large amount of user behavioral data, big data tools can be used for processing.

[0065] In the embodiments of the present disclosure, in the process of determining the resource performance scores for each brush in the corresponding time period according to the behavioral feature data within each time period, since each brush may contain multiple resources, the performance scores can be calculated separately for each resource, and then these scores can be summarized to obtain the overall performance score of the brush.

[0066] In the embodiments of the present disclosure, a positive brush count indicates that the user has a relatively high interest or satisfaction with the resource content in the brush count; a negative brush count indicates that the user has a relatively low interest or satisfaction with the resource content in the brush count. When dividing the brush counts involved in multiple time periods into positive brush counts and negative brush counts according to the resource performance scores of multiple brush counts in multiple time periods, different mechanisms can be adopted for division according to the actual situation. Exemplarily, a division based on a threshold mechanism can be selected, and by comparing with a preset threshold, the positive brush counts and negative brush counts can be divided; similarly, a division based on a quantile mechanism can also be selected, by sorting the resource performance scores, those with higher rankings are divided into positive brush counts, and those with lower rankings are divided into negative brush counts. In particular, as time goes by or the data volume increases, the division criteria for positive and negative brush counts can be dynamically optimized. The above is only an exemplary illustration and does not limit all possible situations for dividing positive brush counts and negative brush counts, but only not exhaustively list them here.

[0067] In the embodiments of the present disclosure, in the process of generating samples for training a resource scoring model according to positive brush counts and negative brush counts, the corresponding features can be first extracted according to the positive brush counts and negative brush counts. Subsequently, corresponding labels are generated for each sample based on the positive brush counts and negative brush counts. Finally, the features and labels of the positive brush counts and negative brush counts are combined to form a training data set.

[0068] In this way, by collecting the behavioral feature data of the target object for the pushed resources in multiple time periods and extracting the resource performance scores based on these behavioral data, the true preferences and interests of the user for the resources can be accurately reflected. By analyzing the positive brush counts and negative brush counts, the behavioral preferences of the target object can be incorporated into the model training, enabling the model to better predict the user's satisfaction and behavioral performance for future resources. By analyzing the user's historical behavior in multiple time periods and constructing training samples using the time series characteristics, the model can be helped to capture the trend of the user's preferences changing over time, so that the prediction ability of the model is not limited to static scenarios, improving the flexibility and prediction ability of the model. Quantifying the user's behavior through the resource performance scores provides a stable basis for the division of positive brush counts and negative brush counts, enhancing the accuracy. Dividing the brush counts into positive brush counts and negative brush counts through the resource performance scores helps to extract effective supervision information from the user's behavior data and can improve the quality of the training samples.

[0069] In some embodiments, according to the behavioral characteristic data within each time period, determining the resource performance score for each brush stroke within the corresponding time period includes: extracting the behavioral characteristics of each brush stroke in different dimensions based on the behavioral characteristic data within each time period, and determining the reference threshold corresponding to the behavioral characteristic of each dimension; comparing the behavioral characteristic of each dimension with the corresponding reference threshold to obtain the comparison result of the behavioral characteristic of each dimension; and determining the resource performance score for each brush stroke according to the comparison results of the behavioral characteristics of multiple dimensions of each brush stroke.

[0070] Here, a dimension refers to different categories or analysis perspectives of user behavioral characteristics, and is used to evaluate the interaction performance of the user with respect to resources from multiple aspects. Each dimension can correspond to a specific behavior, such as a click behavior, a stay behavior, a skip behavior, etc. The analysis results of multiple dimensions contribute to a comprehensive evaluation of the resource performance score.

[0071] Here, the reference threshold is a standard value used to judge the quality of user behavior performance. The thresholds for each dimension may be different and can be set based on historical data or business experience.

[0072] In the embodiments of the present disclosure, the process of extracting the behavioral characteristics of each brush stroke in different dimensions based on the behavioral characteristic data within each time period and determining the reference threshold corresponding to the behavioral characteristic of each dimension can, for multiple brush strokes within each time period, respectively extract the interaction behavior records of the target object with respect to the resource in this brush stroke from the data of each brush stroke, and then respectively extract the corresponding characteristic values according to each predefined dimension. At the same time, historical data can be used as a reference, and the reference threshold for each dimension can be determined through statistical analysis. The mean value can be used as the reference threshold, or the median can be used as the reference threshold. In particular, a fixed reference threshold can also be set according to business objectives or experience, or the reference threshold for each dimension can be recalculated for different time periods. The above is only an exemplary illustration and does not limit all possible situations for determining the reference threshold, and only exhaustive listing is not done here.

[0073] In the embodiments of the present disclosure, the process of comparing the behavioral characteristic of each dimension with the corresponding reference threshold to obtain the comparison result of the behavioral characteristic of each dimension can be implemented using qualitative judgment or quantitative scoring according to requirements. Exemplarily, the result of qualitative judgment is usually "good" or "bad", which is judged according to whether the characteristic value reaches the threshold; the result of quantitative scoring is a numerical value used to represent the relative performance of the characteristic value with respect to the reference threshold and can be calculated through a formula. The above is only an exemplary illustration and does not limit all possible situations of the comparison method and the comparison result, and only exhaustive listing is not done here.

[0074] Here, the resource performance score is a quantitative evaluation result of the overall performance of the user behavior in a brush stroke.

[0075] In the embodiments of the present disclosure, in the process of determining the resource performance score for each brushing based on the comparison results of multiple-dimensional behavioral characteristics of each brushing, weights can be assigned to the behavioral characteristics of each dimension, and weighted summation can be performed according to their importance to generate the final score. Or rules can be defined based on the comparison results of different dimension combinations to directly determine the performance score. In particular, the comparison results of each dimension can be used as input features to train a model to output the resource performance score. The above is only an exemplary illustration and does not limit all possible situations for determining the resource performance score. Only that it is not exhaustively listed here.

[0076] In this way, by decomposing the user's behavioral characteristics into multiple dimensions, a more refined evaluation of the resource performance of each brushing can be carried out. By setting a reference threshold for each dimension and comparing it with the actual user behavior, the specific interaction performance of the user with the resource can be accurately measured, avoiding the evaluation deviation that may be caused by a single-dimension evaluation, and improving the accuracy of the resource performance evaluation. By comprehensively generating the resource performance score based on the comparison results of each dimension, the overall performance of the resource can be more comprehensively reflected, rather than being limited to the performance of only one aspect, improving the comprehensiveness of the resource performance score. By using the user behavior data, setting the reference threshold and performing multi-dimensional comparisons to generate an objective resource performance score, the error of subjective evaluation is avoided.

[0077] In some embodiments, determining the reference threshold corresponding to the behavioral characteristics of each dimension includes: performing statistical analysis on the behavioral characteristics of each dimension to obtain the statistical analysis results of the behavioral characteristics of each dimension; and determining the reference threshold of the behavioral characteristics of each dimension based on the statistical analysis results of the behavioral characteristics of each dimension.

[0078] In the embodiments of the present disclosure, in the process of performing statistical analysis on the behavioral characteristics of each dimension to obtain the statistical analysis results of the behavioral characteristics of each dimension, data distribution analysis can be performed on the behavioral characteristics, so as to further understand the distribution of the behavioral characteristic values of each dimension. Exemplarily, central tendency analysis can be used to measure the central position or typical value of the data distribution, including indicators such as mean, median, and mode. Dispersion analysis can also be used to measure the dispersion or volatility of the data distribution, including indicators such as range, variance, and standard deviation. Analysis methods such as data distribution form and outlier detection can also be used for analysis. The above is only an exemplary illustration and does not limit all possible situations of statistical analysis. Only that it is not exhaustively listed here. Then, corresponding indicators can be extracted based on the data distribution analysis results as the statistical analysis results.

[0079] In the embodiments of the present disclosure, in the process of determining the reference threshold for each dimension behavior feature based on the statistical analysis results of each dimension behavior feature, a fixed value, a percentage, or a range can be selected as the reference threshold according to the statistical analysis results. Exemplarily, when using central tendency analysis, the fixed value of the corresponding analysis index can be used as the reference threshold; when it is necessary to divide different performance levels, an appropriate quantile percentage can be selected as the reference threshold based on the data distribution pattern; when the reference threshold needs to be dynamically set according to the volatility of the data, an appropriate numerical range can be selected as the reference threshold according to the dispersion analysis index. In particular, in some cases, a single statistical index may not be sufficient to determine a reasonable reference threshold, and multiple statistical results can be combined for setting. Further, the statistical analysis results can be used as the basis for setting the reference threshold, and the actual threshold can be adjusted in combination with specific business requirements and scenarios when setting the threshold.

[0080] In this way, by determining the reference threshold through statistical analysis, the threshold can be set based on the real user behavior data and the statistical results such as data distribution, central tendency, and dispersion. This avoids the deviation that may be brought by subjective judgment and empirical setting, and ensures the rationality and accuracy of the threshold. Based on the reference threshold, it can be quickly judged whether the performance of a certain brush in different dimensions meets the expectation, and then a resource performance score can be generated for each brush, providing a direct basis for the optimization, classification, and recommendation of resources.

[0081] In some embodiments, determining the resource performance score for each brush according to the comparison results of the multi-dimensional behavior features of each brush includes: determining the playback performance score of each resource according to the comparison results of the multi-dimensional behavior features of each resource; and determining the resource performance score of each brush according to the playback performance scores of all resources within each brush.

[0082] Here, the playback performance score is an evaluation score for a single resource, reflecting the performance of the resource in multiple user behavior dimensions.

[0083] In the embodiments of the present disclosure, the process of determining the playback performance score of each resource according to the comparison results of the behavior characteristics of each resource in multiple dimensions can quantitatively calculate the overall playback performance of the resource. Exemplarily, a relative threshold scoring method can be adopted. Under each dimension, the difference between the characteristic value of the resource and the corresponding threshold can be calculated. If the characteristic value exceeds the threshold, the score is positive; otherwise, it is negative. This difference is used as the playback performance score under this dimension. A normalized scoring method can also be adopted. Under each dimension, the characteristic value of the resource and the corresponding threshold are normalized proportionally, and the corresponding proportional value is calculated. This proportional value is used as the playback performance score under this dimension. The performance of the characteristic can also be divided into multiple levels, and each level corresponds to a fixed score. Then, under each dimension, the characteristic of the resource is scored, and this score value is used as the playback performance score under this dimension. Next, the overall playback performance scores in all dimensions can be comprehensively calculated, and this score is used as the playback performance score of the resource. Exemplarily, the scores of all dimensions can be weighted and summed according to their weights, and the calculation result is used as the playback performance score of the resource. In particular, if the behavior characteristic data of some dimensions is missing, a default score can be used or the weight of the missing characteristic can be set to 0. The above is only an exemplary illustration and does not limit all possible situations for determining the playback performance score. Only that it is not enumerated here.

[0084] In the embodiments of the present disclosure, the process of determining the resource performance score of each brush stroke according to the playback performance scores of all resources within each brush stroke can aggregate the playback performance scores of multiple resources within the brush stroke, and use the aggregation result as the resource performance score of any brush stroke. Exemplarily, the overall performance of the brush stroke can be reflected by calculating the average value of the playback performance scores of all resources within the brush stroke. Different weights can also be assigned to different resources within the brush stroke, and then the weighted average score is calculated to reflect the differences in the importance of the resources. The maximum or minimum value of the playback performance scores of all resources within the brush stroke can also be taken as the resource performance score of this brush stroke. The above is only an exemplary illustration and does not limit all possible situations for determining the resource performance score of the brush stroke. Only that it is not enumerated here.

[0085] In this way, by first calculating the playback performance score of each resource and then summarizing it to the brush stroke level, the performance of the resource and the overall performance of the brush stroke are separated, forming a hierarchical evaluation process, which avoids the loss of important information that may be caused by directly calculating the brush stroke score, making the evaluation more scientific and clear. By comparing and summarizing multi-dimensional characteristics, the one-sidedness that may be brought by single-dimensional evaluation is avoided, ensuring that the brush stroke score can more accurately represent the true preferences and experiences of users. By first calculating the playback performance score of the resource and then comprehensively calculating the resource performance score of the brush stroke, accurate evaluation can be achieved, which helps with resource screening and brush stroke optimization, making the recommendation strategy more efficient and intelligent.

[0086] In some embodiments, determining the playback performance score of each resource according to the comparison results of the behavioral characteristics of each resource in multiple dimensions includes: obtaining the playback duration of each resource and its reference threshold in the dimension of playback duration; and determining the ratio of the playback duration of each resource to its reference threshold in the dimension of playback duration as the playback performance score of each resource.

[0087] Here, the playback duration refers to the actual viewing time when a user watches a certain resource. Generally, the longer the playback duration, the more interested the user is or the more attractive the resource content is.

[0088] In the embodiments of the present disclosure, the process of obtaining the playback duration of each resource and its reference threshold in the dimension of playback duration may obtain the playback duration data through a user behavior data collection system. Exemplarily, when a user watches a resource on a mobile device or a web page, the client can record the start time and end time of playback in real time and report them to the server. At the same time, the timestamp of the playback request and the timestamp of the end of playback can also be directly recorded by the server, and the time difference is calculated to obtain the playback duration. The above is only an exemplary illustration and does not limit all possible situations for obtaining the playback duration of resources, and only exhaustive listing is not done here. Then, the reference threshold can be calculated from historical data based on the distribution of the playback duration. Exemplarily, the reference threshold of the resource in the dimension of playback duration can be calculated according to the viewing performance of all users for the resource. For example, for a 100-second resource, if the playback duration of more than 80% of the users is greater than a certain duration, the satisfaction threshold of the resource can be set to the corresponding duration. The above is only an exemplary illustration and does not limit all possible situations for obtaining the reference threshold of the playback duration of resources, and only exhaustive listing is not done here.

[0089] In the embodiments of the present disclosure, the process of determining the ratio of the playback duration of each resource to its reference threshold in the dimension of playback duration as the playback performance score of each resource may first calculate the ratio of the playback duration to the reference threshold. Exemplarily, if the playback duration is higher than the reference threshold, the playback performance score is greater than 1, indicating that the performance of the resource is better than the standard. If the playback duration is lower than the reference threshold, the playback performance score is less than 1, indicating that the performance of the resource does not meet the standard. If the playback duration is equal to the reference threshold, the playback performance score is equal to 1, indicating that the performance of the resource just meets the standard. In particular, the playback performance score is generally not limited in range, but it can be restricted or normalized in practical applications.

[0090] In this way, the playback performance of resources can be quantified by the ratio of the playback duration to the reference threshold, replacing subjective judgment in a data-driven manner to ensure more scientific and intuitive comparison between different resources. The reference threshold can be dynamically adjusted according to the resource type or user group to adapt to the evaluation requirements in different scenarios. By calculating the playback performance score in real time, the viewing behavior of users can be quickly reflected, which helps the recommendation system to dynamically adjust the recommendation strategy.

[0091] In some embodiments, according to the resource performance scores of multiple swiping times in multiple time periods, the swiping times involved in multiple time periods are divided into positive swiping times and negative swiping times, including: generating a sequence of swiping performance scores in multiple time periods according to the resource performance scores of multiple swiping times in multiple time periods; traversing the sequence of swiping performance scores, comparing the swiping performance score of each swiping time with the swiping performance score of the adjacent previous swiping time respectively, and taking the swiping time with the increased swiping performance score as the positive swiping time; and taking the swiping time with the decreased swiping performance score as the negative swiping time.

[0092] Here, the sequence of swiping performance scores refers to a set of swiping resource performance scores of swiping times arranged in chronological order, which is used to describe the change trend of swiping performance in multiple time periods. Each element in the sequence represents the comprehensive performance score of a certain swiping time, and the arrangement order of the sequence is consistent with the chronological order. The sequence of swiping performance scores can be used to analyze the performance change trend between swiping times.

[0093] In the embodiments of the present disclosure, the process of generating a sequence of swiping performance scores in multiple time periods according to the resource performance scores of multiple swiping times in multiple time periods may collect the performance scores of all swiping times in multiple time periods, arrange them in chronological order, and then arrange the resource performance scores of each swiping time in chronological order to obtain the sequence of swiping performance scores.

[0094] In the embodiments of the present disclosure, when traversing the sequence of swiping performance scores, starting from the second element of the sequence of swiping performance scores, the scores of the current swiping time and the adjacent previous swiping time are sequentially extracted, and the scores of each pair of adjacent swiping times are compared. Subsequently, if the performance score of the current swiping time is higher than that of the previous swiping time, the current swiping time is taken as the positive swiping time; if the performance score of the current swiping time is lower than that of the previous swiping time, the current swiping time is taken as the negative swiping time.

[0095] In some embodiments, a positive brush count can be understood as the behavior of a user showing positive interest or satisfaction with the resources provided by the recommendation system after performing a refresh operation. For example, the user clicks, browses, favorites, purchases, or gives a positive evaluation, etc. Furthermore, positive samples can be determined based on the resource features and behavior features included in the positive brush count. A negative brush count can be understood as corresponding to the behavior of a user showing negative interest or dissatisfaction with the resources provided by the recommendation system after performing a refresh operation. For example, the user ignores, skips, closes, or gives a negative evaluation, etc. Furthermore, negative samples can be determined based on the resource features and behavior features included in the negative brush count.

[0096] In this way, arranging the brush count performance scores in a sequence according to time can capture the dynamic change trend between brush counts. By comparing brush counts one by one, positive or negative brush counts can be clearly identified. Through the comparison of brush count performance scores, it can be made clear for the recommendation system that the recommendation algorithm or strategy has better effects during certain time periods. Furthermore, the recommendation logic can be adjusted according to the differences in brush count performance, so as to more efficiently adapt to the needs of the user at the current time period.

[0097] Figure 2 The flowchart of the resource scoring method is shown, as Figure 2 shown, the resource scoring method includes:

[0098] S201. Obtain the first resource features, user behavior features, user basic features, and scenario features of the nearest m resources within the current brush count, as well as the second resource features of the candidate resources within the next brush count.

[0099] S202. Input all the features into a Multilayer Perceptron (MLP) for training.

[0100] S203a. Use the MLP to output the first type of score;

[0101] S203b. Use the MLP to output the second type of score.

[0102] Among them, the first type of score can indicate whether the user's behavior performance in the next brush count is better than that in the current brush count.

[0103] Among them, the second type of score can indicate whether the user is satisfied with the candidate resources in the next brush count.

[0104] Here, S203a and S203b can be executed simultaneously.

[0105] Here, the value of m can be set or adjusted according to the recommendation requirements.

[0106] Here, the scenario feature refers to how the user enters or participates in a specific scenario or interface. For example, the user can enter from the home page or from a two-column layout. These different entry paths reflect the diversity of user behavior, product design, and user experience.

[0107] It should be understood that Figure 2 the illustrated schematic diagrams are merely exemplary rather than restrictive, and they are extensible. Those skilled in the art can make various obvious changes and / or substitutions based on Figure 2 the examples, and the resulting technical solutions still fall within the scope of the disclosure of the embodiments of the present disclosure.

[0108] The embodiments of the present disclosure provide a resource scoring method. Figure 3 FIG. is a flowchart of the resource scoring method according to the embodiments of the present disclosure. The resource scoring method can be applied to a resource scoring device. The resource scoring device is located in an electronic device. The electronic device includes but is not limited to a fixed device and / or a mobile device. For example, the fixed device includes but is not limited to a server, and the server can be a cloud server or a general server. For example, the mobile device includes but is not limited to: mobile phones, tablets, personal computers, etc. In some possible implementation manners, the resource scoring method can also be implemented by a processor invoking computer-readable instructions stored in a memory. As Figure 3 shown, the resource scoring method includes:

[0109] Step S301, obtaining the behavior characteristics, basic characteristics, first resource characteristics of the pushed resources in the current time period, and second resource characteristics of the resources to be pushed in the next time period of the target object;

[0110] Step S302, inputting the behavior characteristics, basic characteristics, first resource characteristics, and second resource characteristics into a resource scoring model to obtain a first type of score output by the resource scoring model; wherein, the first type of score is used to predict the relative change in the behavior performance of the target object for the resources to be pushed in the next time period compared to the behavior performance of the target object for the pushed resources in the current time period.

[0111] Here, the current time period refers to the time range currently monitored by the system. The target object refers to the user served by the recommendation system, that is, the individual whose behavior the system hopes to predict and for whom resources are recommended. The behavior characteristics refer to the interaction behavior data of the target object for the pushed resources in the current time period, which can reflect the user's interests and behavior habits in the current time period. The basic characteristics are the static or basic attributes of the target object, which do not change frequently over time and can describe the long-term characteristics of the target object and can be used as an important dimension for predicting user behavior. The first resource characteristics refer to the content characteristics of the pushed resources, which can describe the attributes or labels of these resources.

[0112] Here, the next time period refers to the time range immediately following the end of the current time period. The resource to be pushed refers to the resource that the system plans to recommend to the target object within the next time period. The second resource feature refers to the content characteristics of the resource to be pushed, which is similar to the first resource feature but describes the attributes or tags of the resource planned to be pushed within the next time period. The feature dimensions of the first resource feature and the second resource feature are the same or similar.

[0113] In the embodiments of the present disclosure, the behavioral characteristics of the target object within the current time period can first obtain the user behavior data within the current time period through the log records of the client or the server, and then extract the behavioral characteristics of the target object through the user behavior data.

[0114] In the embodiments of the present disclosure, the basic characteristics of the target object within the current time period can be obtained through the user portrait of the target object. The user portrait is generally formed based on registration information, historical behavior data, or long-term tags and is usually stored in the user attribute database of the recommendation system.

[0115] In the embodiments of the present disclosure, the first resource feature of the resources already pushed within the current time period can be obtained through the recommendation system. The system records each resource pushed to the target object within the current time period and its related attributes, which are usually stored in the log or resource management library of the recommendation system.

[0116] In the embodiments of the present disclosure, the second resource feature of the resources to be pushed within the next time period can be directly extracted according to the content characteristics, tags, scores, etc. of the resources to be pushed. The resources to be pushed are usually selected by the recommendation system from the candidate resource pool, and their characteristics are stored in the resource database.

[0117] In the embodiments of the present disclosure, the process of inputting the behavioral characteristics, basic characteristics, first resource feature, and second resource feature into the resource scoring model to obtain the first type of score output by the resource scoring model can first organize the behavioral characteristics, basic characteristics, first resource feature, and second resource feature into the input format required by the model, and combine the features to construct a complete input vector. Subsequently, the behavioral characteristics and basic characteristics of the target object can be concatenated with the first resource feature and the second resource feature and input as a whole. Then, the concatenated input vector is passed into the resource scoring model, and the model calculates based on the parameters learned from the training data. Finally, the model predicts the change trend of the behavioral performance of the target object towards the resources to be pushed in the next time period relative to the resources already pushed in the current time period through the matching degree between the target object and the resource features.

[0118] In this way, by combining the current behavioral characteristics and basic characteristics of the target object, as well as the historical characteristics (i.e., the first resource characteristics) and future characteristics (i.e., the second resource characteristics) of the resources, the dynamic prediction of user behavior is achieved. By considering multiple characteristics and evaluating the user's potential interest in resources from multiple dimensions, the flexibility and adaptability of the evaluation are improved. By predicting the behavioral changes of the target object in the next time period, the recommendation list can be dynamically adjusted to preferentially push the resources with better prediction performance, improving the accuracy of the recommendation, and thus effectively driving the user's continuous consumption of resources.

[0119] In some embodiments, the resource scoring method further includes: sending a first type of score to a first recommendation model to generate a first recommendation result by the first recommendation model in combination with the first type of score.

[0120] Here, the first recommendation model can be a model for generating a recommendation result according to a given input. The model can be a recommendation algorithm based on machine learning, deep learning, collaborative filtering, causal reasoning, or rule logic. The first recommendation model can be a key component of the recommendation system, responsible for integrating information from multiple dimensions and finally generating a personalized list of recommended resources.

[0121] In the embodiments of the present disclosure, the first type of score can be packaged as a feature vector and provided as an input to the first recommendation model. The first recommendation model receives the first type of score and other feature parameters and generates a final recommendation result according to the logic of the model or the trained rules. Exemplarily, the first recommendation model can combine the first type of score with other feature parameters through a power fusion method to finally generate a first recommendation result.

[0122] In this way, inputting the first type of score into the first recommendation model can further enhance the ability of the recommendation system to capture user preferences, and at the same time, the recommendation strategy can be dynamically adjusted according to the user's behavioral changes to generate personalized recommendation results. By combining the first type of score with the recommendation logic to form a closed loop from behavior analysis to resource recommendation and then to behavior feedback, it helps to improve the accuracy of the recommendation.

[0123] In some embodiments, the resource scoring method further includes: inputting a second type of score generation instruction into a resource scoring model; wherein, the second type of score generation instruction is used to instruct the resource scoring model to generate a second type of score; wherein, the second type of score is used to predict the satisfaction of the target object with the resources to be pushed in the next time period; obtaining the second type of score of the resources to be pushed output by the resource scoring model based on the first resource characteristics, the second resource characteristics, the behavioral characteristics, and the basic characteristics.

[0124] Here, the second type of scoring generation instruction is a control instruction used to trigger the resource scoring model to generate the second type of score. This instruction instructs the resource scoring model to calculate the predicted satisfaction score of the target object for the resource to be pushed, that is, the second type of score, using the first resource feature, the second resource feature, the behavior feature, and the basic feature.

[0125] In the embodiments of the present disclosure, in the process of inputting the second type of scoring generation instruction into the resource scoring model, the second type of scoring generation instruction can be sent to the model together with the input features. Exemplarily, this instruction can include the type of score to be generated.

[0126] In the embodiments of the present disclosure, first, the first resource feature, the second resource feature, the behavior feature, and the basic feature are input into the resource scoring model. Subsequently, the model predicts the satisfaction of the target object for the resource to be pushed by combining these input features according to the weights or rules learned from the training data.

[0127] In this way, by generating the second type of score, the system can screen out resources that not only have good user behavior performance but also that the user may be truly interested in and satisfied with. Since the generation process of the second type of score takes into account the user's personalized features and resource features, the system can accurately predict the user's satisfaction with the resources based on the score and preferentially recommend resources with high scores.

[0128] In some embodiments, the resource scoring method further includes: sending the first type of score and the second type of score to the second recommendation model, so that the second recommendation model generates a second recommendation result by combining the first type of score and the second type of score.

[0129] Here, the second recommendation model can be used to calculate the final recommendation result of the resource to be pushed according to the input first type of score, the second type of score, and other possible features. The second recommendation model can be a key component of the recommendation system, responsible for integrating information from multiple dimensions and finally generating a personalized recommended resource list.

[0130] In the embodiments of the present disclosure, in the process of sending the first type of score and the second type of score to the second recommendation model, the first type of score and the second type of score calculated by the resource scoring model can be obtained first. Subsequently, the first type of score and the second type of score of each resource to be pushed are constructed into an input format recognizable by the model and then the scores are sent to the second recommendation model.

[0131] In the embodiments of the present disclosure, after receiving the first type of score and the second type of score of the resource to be pushed, the second recommendation model can calculate the comprehensive score of each resource, and then sort all the resources to be pushed according to the comprehensive score, and select the resource with the highest score as the recommendation result.

[0132] Exemplarily, the second recommendation model can combine the first type of scores, the second type of scores, and other feature parameters through a power fusion method, and finally generate a second recommendation result.

[0133] In this way, by combining the first type of scores and the second type of scores, the second recommendation model can more comprehensively evaluate the recommendation value of resources, and can significantly improve the accuracy of the recommendation result and user satisfaction. By combining the first type of scores and the second type of scores, it can not only provide resources that the user may click on, but also provide content that the user is truly interested in and satisfied with at the current time period, effectively driving the user's continuous consumption of resources.

[0134] Embodiments of the present disclosure provide a training device for a resource scoring model, as Figure 4 shown. The device may include: a first acquisition module 401, configured to acquire the first resource features of the resources already pushed within the first time period and the second resource features of the resources to be pushed within the second time period, where the second time period is the next time period after the first time period; a second acquisition module 402, configured to acquire the behavior features of the target object for the resources already pushed within the first time period and the basic features of the target object; a training module 403, configured to input the first resource features, the second resource features, the behavior features, and the basic features into the resource scoring model, so that the resource scoring model outputs at least the first type of scores for the resources to be pushed; where the first type of scores is used to predict the relative change in the behavior performance of the target object for the resources to be pushed within the second time period compared to the behavior performance of the resources already pushed within the first time period.

[0135] In some embodiments, the device further includes: an expansion module, configured to expand the output ability of the resource scoring model, so that the resource scoring model additionally outputs the second type of scores for the resources to be pushed based on the first resource features, the second resource features, the behavior features, and the basic features; where the second type of scores is used to predict the satisfaction of the target object with the resources to be pushed within the second time period.

[0136] In some embodiments, the device further includes: a data collection module, configured to collect the behavior feature data of the target object for the resources already pushed within multiple time periods; a score determination module, configured to determine the resource performance scores for each brush within the corresponding time period according to the behavior feature data within each time period; where each time period includes at least one brush; a brush division module, configured to divide the brushes involved within multiple time periods into positive brushes and negative brushes according to the resource performance scores of multiple brushes within multiple time periods; a sample generation module, configured to generate samples for training the resource scoring model according to the positive brushes and the negative brushes.

[0137] In some embodiments, the score determination module includes: a threshold determination sub-module, configured to extract the behavior characteristics of each brushing in different dimensions according to the behavior characteristic data within each time period, and determine the reference threshold corresponding to the behavior characteristics of each dimension; a threshold comparison sub-module, configured to compare the behavior characteristics of each dimension with the corresponding reference threshold to obtain the comparison result of the behavior characteristics of each dimension; and a score determination sub-module, configured to determine the resource performance score of each brushing according to the comparison results of the behavior characteristics of multiple dimensions of each brushing.

[0138] In some embodiments, the threshold determination sub-module is further configured to: perform statistical analysis on the behavior characteristics of each dimension to obtain the statistical analysis result of the behavior characteristics of each dimension; and determine the reference threshold of the behavior characteristics of each dimension based on the statistical analysis result of the behavior characteristics of each dimension.

[0139] In some embodiments, the score determination sub-module is further configured to: determine the playback performance score of each resource according to the comparison results of the behavior characteristics of multiple dimensions of each resource; and determine the resource performance score of each brushing according to the playback performance scores of all resources within each brushing.

[0140] In some embodiments, the score determination sub-module is further configured to: obtain the playback duration of each resource and its reference threshold in the playback duration dimension; and determine the ratio of the playback duration of each resource to its reference threshold in the playback duration dimension as the playback performance score of each resource.

[0141] In some embodiments, the brushing division module is further configured to: generate a brushing performance score sequence within multiple time periods according to the resource performance scores of multiple brushings within multiple time periods; traverse the brushing performance score sequence, compare the brushing performance score of each brushing with the brushing performance score of the adjacent previous brushing respectively, and regard the brushing with an increasing brushing performance score as a positive brushing; and regard the brushing with a decreasing brushing performance score as a negative brushing.

[0142] For the specific functions and example descriptions of the modules and sub-modules of the training device of the resource scoring model according to the embodiments of the present disclosure, reference may be made to the relevant descriptions of the corresponding steps in the above method embodiments, which will not be elaborated herein.

[0143] The training device of the resource scoring model according to the embodiments of the present disclosure, through the fusion of multi-dimensional features, enables the model to comprehensively consider the complex relationships among the user's historical behaviors, user attributes, and resource attributes, and thus can more accurately evaluate which resources are most suitable for the current user state and make more reliable predictions about future user behaviors.

[0144] The embodiments of the present disclosure provide a resource scoring device, such as Figure 5As shown in the figure, the device may include: a third acquisition module 501, configured to acquire the behavior characteristics, basic characteristics, first resource characteristics of the pushed resources, and second resource characteristics of the resources to be pushed in the next time period of the target object within the current time period; a resource scoring module 502, configured to input the behavior characteristics, basic characteristics, first resource characteristics, and second resource characteristics into a resource scoring model, and obtain a first type of score output by the resource scoring model; wherein, the first type of score is used to predict the relative change in the behavior performance of the target object for the resources to be pushed in the next time period compared to the behavior performance of the target object for the pushed resources in the current time period.

[0145] In some embodiments, the device further includes: a first result generation module, configured to send the first type of score to a first recommendation model, so that the first recommendation model generates a first recommendation result in combination with the first type of score.

[0146] In some embodiments, the device further includes: an instruction generation module, configured to input a second type of score generation instruction into the resource scoring model; wherein, the second type of score generation instruction is used to instruct the resource scoring model to generate a second type of score; wherein, the second type of score is used to predict the satisfaction degree of the target object for the resources to be pushed in the next time period; a second score acquisition module, configured to acquire the second type of score of the resources to be pushed output by the resource scoring model based on the first resource characteristics, second resource characteristics, behavior characteristics, and basic characteristics.

[0147] In some embodiments, the device further includes: a second result generation module, configured to send the first type of score and the second type of score to a second recommendation model, so that the second recommendation model generates a second recommendation result in combination with the first type of score and the second type of score.

[0148] For the specific functions and example descriptions of each module and sub-module of the resource scoring device according to the embodiments of the present disclosure, reference may be made to the relevant descriptions of the corresponding steps in the above method embodiments, which will not be elaborated herein.

[0149] The resource scoring device according to the embodiments of the present disclosure realizes dynamic prediction of user behavior by combining the current behavior characteristics and basic characteristics of the target object, as well as the resource characteristics in the current time period and the resource characteristics in the next time period. By considering various characteristics and evaluating the potential interest of users in resources from multiple dimensions, the flexibility and adaptability are improved. By predicting the behavior changes of the target object in the next time period, the recommendation list can be dynamically adjusted, and resources with better predicted performance are preferentially pushed, effectively driving the continuous consumption of resources by users.

[0150] The embodiments of the present disclosure provide a schematic diagram of a scenario for a training method of a resource scoring model, as Figure 6 shown.

[0151] As described above, the training method of the resource scoring model provided by the embodiments of the present disclosure is applied to an electronic device. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers.

[0152] Specifically, the electronic device can specifically perform the following operations:

[0153] Obtain the first resource features of the resources that have been pushed within the first time period and the second resource features of the resources to be pushed within the second time period, where the second time period is the next time period after the first time period;

[0154] Obtain the behavioral features of the target object for the resources that have been pushed and the basic features of the target object within the first time period;

[0155] Input the first resource features, the second resource features, the behavioral features, and the basic features into the resource scoring model so that the resource scoring model outputs at least the first type of score for the resources to be pushed; wherein, the first type of score is used to predict the relative change in the behavioral performance of the target object for the resources to be pushed within the second time period compared to the behavioral performance for the resources that have been pushed within the first time period.

[0156] The embodiments of the present disclosure provide a schematic diagram of the scenario of a resource scoring method, as Figure 7 shown.

[0157] As described above, the resource scoring method provided by the embodiments of the present disclosure is applied to an electronic device. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers.

[0158] Specifically, the electronic device can specifically perform the following operations:

[0159] Obtain the behavioral features of the target object, the basic features, the first resource features of the resources that have been pushed, and the second resource features of the resources to be pushed within the next time period during the current time period;

[0160] Input the behavioral features, the basic features, the first resource features, and the second resource features into the resource scoring model to obtain the first type of score output by the resource scoring model; wherein, the first type of score is used to predict the relative change in the behavioral performance of the target object for the resources to be pushed within the next time period compared to the behavioral performance of the target object for the resources that have been pushed during the current time period.

[0161] It should be understood that Figure 6 and Figure 7 the scenario diagrams shown are merely illustrative and not restrictive. Those skilled in the art can be based on Figure 6 andFigure 7 Various obvious changes and / or substitutions are made to the examples, and the resulting technical solutions still fall within the scope of the disclosure of the embodiments of the present disclosure.

[0162] In the technical solutions of the present disclosure, the acquisition, storage, and application of the user's personal information involved all comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0163] According to the embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0164] Figure 8 FIG. shows a schematic block diagram of an exemplary electronic device 800 that can be used to implement the embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as, for example, a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, for example, a personal digital assistant, a cellular phone, a smartphone, a wearable device, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely exemplary and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0165] As Figure 8 shown, the device 800 includes a computing unit 801, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the device 800 can also be stored. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0166] A plurality of components in the device 800 are connected to the I / O interface 805, including: an input unit 806, such as a keyboard, a mouse, etc.; an output unit 807, such as various types of displays, speakers, etc.; a storage unit 808, such as a disk, an optical disc, etc.; and a communication unit 809, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 809 allows the device 800 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0167] The computing unit 801 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various dedicated Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 executes the various methods and processes described above, such as the training method of the resource scoring model and the resource scoring method. For example, in some embodiments, the training method of the resource scoring model and the resource scoring method can be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as the storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded into the RAM 803 and executed by the computing unit 801, one or more steps of the training method of the resource scoring model and the resource scoring method described above can be executed. Alternatively, in other embodiments, the computing unit 801 can be configured to execute the training method of the resource scoring model and the resource scoring method by any other suitable means (e.g., by means of firmware).

[0168] The various embodiments of the systems and techniques described above in this specification can be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application-specific standard products (ASSPs), system on chip (SOC) systems, complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which may be a special-purpose or general-purpose programmable processor that receives data and instructions from, and transmits data and instructions to, a storage system, at least one input device, and at least one output device.

[0169] The program code for implementing the methods of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions / operations specified in the flowchart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

[0170] In the context of this disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0171] To provide for interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a cathode ray tube (CRT) or a liquid crystal display (LCD) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can also be used to provide for interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0172] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0173] A computer system may include a client and a server. The client and the server are generally far away from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the respective computers and having a client-server relationship with each other. The server may be a cloud server, or a server of a distributed system, or a server incorporating a blockchain.

[0174] It should be understood that various forms of the processes shown above may be used, steps may be reordered, added or deleted. For example, the steps recited in the present disclosure may be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved, and no limitations are imposed herein.

[0175] The above specific embodiments do not constitute a limitation on the protection scope of the present disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the principles of the present disclosure shall be included within the protection scope of the present disclosure.

Claims

1. A method for training a resource scoring model, comprising: Acquire a first resource feature of a resource that has been pushed in a first time period and a second resource feature of a resource to be pushed in a second time period, wherein the second time period is a time period next to the first time period; Acquire the behavior characteristics of the target object with respect to the pushed resource and the basic characteristics of the target object in the first time period; The first resource feature, the second resource feature, the behavior feature and the basic feature are input into a resource scoring model so that the resource scoring model outputs at least a first type of score for the resource to be pushed; wherein the first type of score is used to predict the relative change of the target object's behavior performance towards the resource to be pushed during the second time period and the behavior performance towards the pushed resource during the first time period.

2. The method according to claim 1, wherein: The method further comprises: Expand the output capability of the resource scoring model so that the resource scoring model additionally outputs a second type of score for the resource to be pushed based on the first resource feature, the second resource feature, the behavioral feature, and the basic feature; wherein the second type of score is used to predict the target object's satisfaction with the resource to be pushed within the second time period.

3. The method according to claim 1 or 2, wherein: The method further comprises: Collecting behavior feature data of the target object for pushed resources in multiple time periods; Determine the resource performance score of each brush in the corresponding time period according to the behavior characteristic data in each time period; wherein each time period includes at least one brush; According to the resource performance scores of the multiple brushing times in the multiple time periods, the brushing times involved in the multiple time periods are divided into positive brushing times and negative brushing times; According to the positive brush times and the negative brush times, samples for training the resource scoring model are generated.

4. The method according to claim 3, wherein: Determining the resource performance score of each refresh in the corresponding time period according to the behavior feature data in each time period includes: According to the behavior feature data in each time period, the behavior features of each swipe in different dimensions are extracted, and the reference threshold corresponding to the behavior features in each dimension is determined; Compare each dimensional behavior feature with the corresponding reference threshold to obtain a comparison result of each dimensional behavior feature; The resource performance score of each brush is determined based on the comparison results of the multiple dimensional behavioral characteristics of each brush.

5. The method according to claim 4, wherein: Determining the reference threshold corresponding to each dimensional behavior feature includes: Perform statistical analysis on the behavioral characteristics of each dimension to obtain statistical analysis results of the behavioral characteristics of each dimension; Based on the statistical analysis results of the behavioral characteristics of each dimension, a reference threshold of the behavioral characteristics of each dimension is determined.

6. The method according to claim 4, wherein: Determining the resource performance score of each brushing time according to the comparison results of the multiple dimensional behavior characteristics of each brushing time includes: Determine the playback performance score of each resource based on the comparison results of the behavioral characteristics of each resource in multiple dimensions; The resource performance score of each refresh is determined based on the playback performance scores of all resources in each refresh.

7. The method according to claim 6, wherein: Determining the playback performance score of each resource according to the comparison results of the behavior characteristics of each resource in multiple dimensions includes: Get the playback duration of each resource and its reference threshold in the playback duration dimension; The ratio of the playback duration of each resource to its reference threshold in the playback duration dimension is determined as the playback performance score of each resource.

8. The method according to claim 3, wherein: The dividing the brushing times involved in the multiple time periods into positive brushing times and negative brushing times according to the resource performance scores of the multiple brushing times in the multiple time periods includes: Generating a sequence of swipe performance scores within the multiple time periods according to the resource performance scores of the multiple swipes within the multiple time periods; The brush performance score sequence is traversed, and the brush performance score of each brush is compared with the adjacent previous brush, and the brush performance score of the brush performance score is increased as the positive brush; and the brush performance score of the brush performance score is reduced as the negative brush.

9. A resource scoring method, comprising: Obtaining the behavior characteristics, basic characteristics, first resource characteristics of pushed resources, and second resource characteristics of resources to be pushed in the next time period of the target object in the current time period; The behavioral features, the basic features, the first resource features and the second resource features are input into a resource scoring model to obtain a first type of score output by the resource scoring model; wherein the first type of score is used to predict the relative change in the behavioral performance of the target object towards the resource to be pushed in the next time period and the behavioral performance of the target object towards the pushed resource in the current time period; wherein the resource scoring model is obtained by training the method described in any one of claims 1 to 8.

10. The method according to claim 9, wherein: The method further comprises: The first category score is sent to a first recommendation model, so that the first recommendation model combines the first category score with the first category score to generate a first recommendation result.

11. The method according to claim 9, wherein: The method further comprises: Inputting a second type of score generation instruction into the resource scoring model; wherein the second type of score generation instruction is used to instruct the resource scoring model to generate a second type of score; wherein the second type of score is used to predict the target object's satisfaction with the resource to be pushed in the next time period; A second type of score for the resource to be pushed output by the resource scoring model based on the first resource feature, the second resource feature, the behavior feature and the basic feature is obtained.

12. The method according to claim 11, wherein: The method further comprises: The first category score and the second category score are sent to a second recommendation model, so that the second recommendation model combines the first category score and the second category score to generate a second recommendation result.

13. A training device for a resource scoring model, comprising: A first acquisition module, used to acquire a first resource feature of a resource that has been pushed in a first time period and a second resource feature of a resource to be pushed in a second time period, wherein the second time period is a time period next to the first time period; A second acquisition module is used to acquire the behavior characteristics of the target object on the pushed resource and the basic characteristics of the target object in the first time period; A training module is used to input the first resource feature, the second resource feature, the behavior feature and the basic feature into a resource scoring model so that the resource scoring model at least outputs a first type of score for the resource to be pushed; wherein the first type of score is used to predict the relative change of the target object's behavior performance towards the resource to be pushed during the second time period and the behavior performance towards the pushed resource during the first time period.

14. A resource scoring device, comprising: The third acquisition module is used to acquire the behavior characteristics, basic characteristics, first resource characteristics of pushed resources, and second resource characteristics of resources to be pushed in the next time period of the target object in the current time period; A resource scoring module, used to input the behavioral features, the basic features, the first resource features and the second resource features into a resource scoring model to obtain a first type of score output by the resource scoring model; wherein the first type of score is used to predict the relative change of the target object's behavioral performance towards the resource to be pushed in the next time period and the target object's behavioral performance towards the pushed resource in the current time period; wherein the resource scoring model is obtained by training the method described in any one of claims 1 to 8.

15. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to at least one processor; wherein, The memory stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 12.

16. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are for causing a computer to perform a method according to any one of claims 1-12.

17. A computer program product comprising a computer program stored on a storage medium, the computer program implementing the method according to any one of claims 1 to 12 when executed by a processor.

Images